Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,184 +1,100 @@
|
|
1 |
-
import gradio as gr
|
2 |
import os
|
3 |
import torch
|
|
|
|
|
4 |
from transformers import AutoProcessor, AutoModel
|
5 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
6 |
from langchain_community.vectorstores import FAISS
|
7 |
from langchain.chains import RetrievalQA
|
8 |
from langchain_community.llms import HuggingFacePipeline
|
9 |
-
from PIL import Image
|
10 |
from langchain_community.document_loaders import PyPDFLoader
|
11 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
12 |
|
|
|
|
|
13 |
class MultimodalRAG:
|
14 |
-
def __init__(self, pdf_path
|
15 |
self.processor = AutoProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
16 |
self.vision_model = AutoModel.from_pretrained("openai/clip-vit-base-patch32")
|
17 |
self.text_embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
18 |
-
|
19 |
-
self.documents =
|
20 |
-
self.vector_store =
|
21 |
-
|
22 |
-
self.qa_chain = None
|
23 |
-
|
24 |
try:
|
25 |
self.llm = HuggingFacePipeline.from_model_id(
|
26 |
model_id="google/flan-t5-large",
|
27 |
task="text2text-generation",
|
28 |
-
model_kwargs={"temperature": 0.7, "max_length": 512}
|
29 |
)
|
30 |
-
except Exception
|
31 |
-
print(f"Error loading flan-t5 model: {e}")
|
32 |
from langchain.llms import OpenAI
|
33 |
self.llm = OpenAI(temperature=0.7)
|
34 |
-
|
35 |
-
if pdf_path and os.path.exists(pdf_path):
|
36 |
-
self.load_pdf(pdf_path)
|
37 |
-
|
38 |
-
def load_pdf(self, pdf_path):
|
39 |
-
if not os.path.exists(pdf_path):
|
40 |
-
raise FileNotFoundError(f"PDF file not found: {pdf_path}")
|
41 |
-
|
42 |
-
loader = PyPDFLoader(pdf_path)
|
43 |
-
self.documents = loader.load()
|
44 |
|
45 |
-
text_splitter = RecursiveCharacterTextSplitter(
|
46 |
-
chunk_size=1000,
|
47 |
-
chunk_overlap=200
|
48 |
-
)
|
49 |
-
self.documents = text_splitter.split_documents(self.documents)
|
50 |
-
|
51 |
-
self.vector_store = FAISS.from_documents(self.documents, self.text_embeddings)
|
52 |
-
|
53 |
self.retriever = self.vector_store.as_retriever(search_kwargs={"k": 2})
|
54 |
-
|
55 |
self.qa_chain = RetrievalQA.from_chain_type(
|
56 |
llm=self.llm,
|
57 |
chain_type="stuff",
|
58 |
retriever=self.retriever,
|
59 |
return_source_documents=True
|
60 |
)
|
61 |
-
|
62 |
-
return f"Successfully loaded and processed PDF: {pdf_path}"
|
63 |
|
64 |
-
def
|
65 |
-
|
66 |
-
|
67 |
-
|
|
|
68 |
|
69 |
-
|
|
|
70 |
inputs = self.processor(images=image, return_tensors="pt")
|
71 |
with torch.no_grad():
|
72 |
-
|
73 |
-
return image_features
|
74 |
-
|
75 |
-
def generate_image_description(self, image_features):
|
76 |
-
return "a photo"
|
77 |
-
|
78 |
-
def retrieve_related_documents(self, query_text, image_path=None):
|
79 |
-
if image_path:
|
80 |
-
image_features = self.process_image(image_path)
|
81 |
-
|
82 |
-
if image_features is not None:
|
83 |
-
image_query = self.generate_image_description(image_features)
|
84 |
-
|
85 |
-
enhanced_query = f"{query_text} {image_query}"
|
86 |
-
else:
|
87 |
-
enhanced_query = query_text
|
88 |
-
else:
|
89 |
-
enhanced_query = query_text
|
90 |
|
91 |
-
|
92 |
-
return
|
93 |
|
94 |
def answer_query(self, query_text, image_path=None):
|
95 |
-
if not self.vector_store or not self.qa_chain:
|
96 |
-
return "Please upload a PDF document first."
|
97 |
-
|
98 |
if image_path:
|
99 |
-
|
|
|
|
|
100 |
else:
|
101 |
-
|
102 |
|
103 |
-
result = self.qa_chain({"query":
|
104 |
-
|
105 |
answer = result["result"]
|
106 |
-
sources = [doc.
|
107 |
-
|
108 |
return answer, sources
|
109 |
|
110 |
-
rag_system = MultimodalRAG()
|
111 |
|
112 |
-
def
|
113 |
-
if pdf_file is None:
|
114 |
-
return "No file uploaded"
|
115 |
-
|
116 |
-
file_path = pdf_file.name
|
117 |
-
try:
|
118 |
-
result = rag_system.load_pdf(file_path)
|
119 |
-
return result
|
120 |
-
except Exception as e:
|
121 |
-
return f"Error processing PDF: {str(e)}"
|
122 |
-
|
123 |
-
def save_image(image):
|
124 |
-
if image is None:
|
125 |
-
return None
|
126 |
-
|
127 |
-
temp_path = "temp_image.jpg"
|
128 |
-
image.save(temp_path)
|
129 |
-
return temp_path
|
130 |
-
|
131 |
-
def process_query(query, pdf_file, image=None):
|
132 |
-
if not query.strip():
|
133 |
-
return "Please enter a question", []
|
134 |
-
|
135 |
if pdf_file is None:
|
136 |
-
return "Please upload a PDF
|
137 |
-
|
|
|
138 |
image_path = None
|
139 |
-
if
|
140 |
-
image_path =
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
upload_button = gr.Button("Process PDF")
|
161 |
-
status_output = gr.Textbox(label="Status")
|
162 |
-
|
163 |
-
upload_button.click(
|
164 |
-
fn=upload_pdf,
|
165 |
-
inputs=[pdf_input],
|
166 |
-
outputs=[status_output]
|
167 |
-
)
|
168 |
-
|
169 |
-
with gr.Column(scale=2):
|
170 |
-
image_input = gr.Image(label="Optional: Upload an Image", type="pil")
|
171 |
-
query_input = gr.Textbox(label="Ask a question")
|
172 |
-
submit_button = gr.Button("Submit Question")
|
173 |
-
|
174 |
-
answer_output = gr.Textbox(label="Answer")
|
175 |
-
sources_output = gr.JSON(label="Sources")
|
176 |
-
|
177 |
-
submit_button.click(
|
178 |
-
fn=process_query,
|
179 |
-
inputs=[query_input, pdf_input, image_input],
|
180 |
-
outputs=[answer_output, sources_output]
|
181 |
-
)
|
182 |
|
183 |
if __name__ == "__main__":
|
184 |
-
|
|
|
|
|
1 |
import os
|
2 |
import torch
|
3 |
+
from PIL import Image
|
4 |
+
import gradio as gr
|
5 |
from transformers import AutoProcessor, AutoModel
|
6 |
from langchain_community.embeddings import HuggingFaceEmbeddings
|
7 |
from langchain_community.vectorstores import FAISS
|
8 |
from langchain.chains import RetrievalQA
|
9 |
from langchain_community.llms import HuggingFacePipeline
|
|
|
10 |
from langchain_community.document_loaders import PyPDFLoader
|
11 |
from langchain.text_splitter import RecursiveCharacterTextSplitter
|
12 |
|
13 |
+
os.environ["CUDA_VISIBLE_DEVICES"] = ""
|
14 |
+
|
15 |
class MultimodalRAG:
|
16 |
+
def __init__(self, pdf_path):
|
17 |
self.processor = AutoProcessor.from_pretrained("openai/clip-vit-base-patch32")
|
18 |
self.vision_model = AutoModel.from_pretrained("openai/clip-vit-base-patch32")
|
19 |
self.text_embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
|
20 |
+
|
21 |
+
self.documents = self._load_and_split(pdf_path)
|
22 |
+
self.vector_store = FAISS.from_documents(self.documents, self.text_embeddings)
|
23 |
+
|
|
|
|
|
24 |
try:
|
25 |
self.llm = HuggingFacePipeline.from_model_id(
|
26 |
model_id="google/flan-t5-large",
|
27 |
task="text2text-generation",
|
28 |
+
model_kwargs={"temperature": 0.7, "max_length": 512, "device": -1}
|
29 |
)
|
30 |
+
except Exception:
|
|
|
31 |
from langchain.llms import OpenAI
|
32 |
self.llm = OpenAI(temperature=0.7)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
self.retriever = self.vector_store.as_retriever(search_kwargs={"k": 2})
|
|
|
35 |
self.qa_chain = RetrievalQA.from_chain_type(
|
36 |
llm=self.llm,
|
37 |
chain_type="stuff",
|
38 |
retriever=self.retriever,
|
39 |
return_source_documents=True
|
40 |
)
|
|
|
|
|
41 |
|
42 |
+
def _load_and_split(self, pdf_path):
|
43 |
+
loader = PyPDFLoader(pdf_path)
|
44 |
+
docs = loader.load()
|
45 |
+
splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
46 |
+
return splitter.split_documents(docs)
|
47 |
|
48 |
+
def _get_image_features(self, image_path):
|
49 |
+
image = Image.open(image_path).convert("RGB")
|
50 |
inputs = self.processor(images=image, return_tensors="pt")
|
51 |
with torch.no_grad():
|
52 |
+
return self.vision_model.get_image_features(**inputs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
+
def _generate_image_description(self, image_features):
|
55 |
+
return "an image"
|
56 |
|
57 |
def answer_query(self, query_text, image_path=None):
|
|
|
|
|
|
|
58 |
if image_path:
|
59 |
+
feats = self._get_image_features(image_path)
|
60 |
+
img_desc = self._generate_image_description(feats)
|
61 |
+
full_query = f"{query_text} {img_desc}"
|
62 |
else:
|
63 |
+
full_query = query_text
|
64 |
|
65 |
+
result = self.qa_chain({"query": full_query})
|
|
|
66 |
answer = result["result"]
|
67 |
+
sources = [doc.metadata for doc in result.get("source_documents", [])]
|
|
|
68 |
return answer, sources
|
69 |
|
|
|
70 |
|
71 |
+
def run_rag(pdf_file, query, image_file=None):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
if pdf_file is None:
|
73 |
+
return "Please upload a PDF.", []
|
74 |
+
|
75 |
+
pdf_path = pdf_file.name
|
76 |
image_path = None
|
77 |
+
if image_file:
|
78 |
+
image_path = image_file.name
|
79 |
+
|
80 |
+
rag = MultimodalRAG(pdf_path)
|
81 |
+
answer, sources = rag.answer_query(query, image_path)
|
82 |
+
return answer, sources
|
83 |
+
|
84 |
+
iface = gr.Interface(
|
85 |
+
fn=run_rag,
|
86 |
+
inputs=[
|
87 |
+
gr.File(label="PDF Document", file_types=[".pdf"]),
|
88 |
+
gr.Textbox(label="Query", placeholder="Enter your question here..."),
|
89 |
+
gr.File(label="Optional Image", file_types=[".png", ".jpg", ".jpeg"], optional=True)
|
90 |
+
],
|
91 |
+
outputs=[
|
92 |
+
gr.Textbox(label="Answer"),
|
93 |
+
gr.JSON(label="Source Documents")
|
94 |
+
],
|
95 |
+
title="Multimodal RAG QA",
|
96 |
+
description="Upload a PDF, ask a question, optionally provide an image."
|
97 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
99 |
if __name__ == "__main__":
|
100 |
+
iface.launch(server_name="0.0.0.0", server_port=7860)
|